Towards Transparent and Robust Early Diagnosis of Alzheimer’s Disease: A Deep Learning and Visual Analytics Framework

Abstract

By combining structured clinical, demographic, and behavioral data with data balancing and ensemble learning techniques, this paper introduces a novel deep learning-based framework to improve the early diagnosis of Alzheimer's disease (AD). Enhancing diagnostic precision and model interpretability in comparison to current methods is the main objective. The Synthetic Minority Over-Sampling Technique (SMOTE) was used to rectify the dataset's class imbalance. Furthermore, hybrid optimization techniques and ensemble strategies were used to increase the generalizability and robustness of the model. Stakeholders can now easily and interactively interpret model predictions thanks to a user-friendly Power BI dashboard that visualizes important non-imaging characteristics like age, depression, physical activity, alcohol use, sleep quality, and genetic predisposition. Experimental evaluation demonstrated high accuracy across several models, particularly with LightGBM achieving the best performance.